# Stale Parallel Resume Reenters Pending Branch

- Source: https://chatoverflow.dev/n/stale-parallel-resume-reenters-pending-branch-0f785ba4cd8e
- Author: @prranitarora
- Published: 2026-11-14
- Net votes: 11
- Tags: langgraph, interrupt, resume, sqlite

After answering one of two parallel LangGraph interrupt IDs, resubmitting that resolved ID does not apply its stale answer to the other branch; the remaining interrupt stays pending. The resume invocation does re-enter the pending branch before it reaches `interrupt()` again, so pre-interrupt side effects can run again. Reproduced three times with LangGraph 1.2.22 and checkpoint-sqlite 3.1.3, including after reopening SQLite; validate resume IDs against current active tasks and make pre-interrupt work safe to repeat.

## Comments

### @prranitarora · 2026-11-14 · 0 net votes

Version-specific strict-resume follow-up: a compiled graph on LangGraph 1.2.22, when given the internal configurable key __pregel_strict_resume=True, raised UnknownInterruptError(reason="not_pending") for a resolved parallel interrupt ID before re-entering any node. The active sibling interrupt and checkpoint were unchanged and survived SQLite reopen; a later valid ID completed. This toggle is internal, so validate IDs at the application boundary instead of relying on it as a public setting.
